A Hybrid Approach Based on Collaborative Filtering to Recommending Mobile Apps

Abstract

With the rapid emergence of mobile devices, smart phones have penetrated into every aspect of people's daily life. The explosive growth of mobile applications makes it difficult for mobile users to find suitable and interesting applications. Mobile app recommendation has been explored by many researchers and some industry solutions are proposed for mobile users. Collaborative filter (CF) is a popular technique in recommendation system, but it requires explicit feedback data like ratings, which are usually difficult to be collected. Many application markets provide keywords search functions and recommend applications with high download counts. However, downloading an application is a vague indicator of whether the user truly likes that application, as the user may probably uninstall that application immediately after it has been installed. Some other industry solutions involve users' personal data such as social networking, which may result in privacy leaks. In this paper, we propose two hybrid models based on collaborative filtering to make mobile app recommendations. We leverage RFD (Recency, Frequency, Duration) model to label the preference for users over applications from users' usage data. The first model, namely, improved item-oriented collaborative filtering (IIOCF), improves the performance of the item-oriented approach by leveraging the latent factor model to discover latent factors among applications. The second model (HLF) is a hybrid model of the latent factor and item-oriented approach. This model sums the predictions of the latent factor model and item-oriented approach, thereby capturing the advantages of both approaches. Our experiment results over 6,568 applications and 25,302 users clearly show that HLF model has better performance than both the item-oriented and latent factor approach.